Spaces:
Sleeping
Sleeping
File size: 9,223 Bytes
16a677c 938aa43 16a677c 938aa43 16a677c 938aa43 16a677c 938aa43 941e967 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 | #!/usr/bin/env python3
"""
Sismik Sınıflandırma - HuggingFace Space (Gradio)
===================================================
RandomForest modeliyle MSEED dosyalarını DEPREM / NOISE olarak sınıflandırır.
HF Spaces kurulumu:
1. huggingface.co/new-space -> SDK: Gradio, Hardware: CPU basic (ücretsiz)
2. Bu dosyayı app.py olarak yükle
3. requirements.txt'i yükle
4. rf_model.pkl dosyasını da Space'e yükle (aynı dizine)
"""
import os
import json
import tempfile
import warnings
warnings.filterwarnings("ignore")
import numpy as np
import joblib
import gradio as gr
from obspy import read
from obspy.signal.trigger import classic_sta_lta, trigger_onset
from obspy.signal.filter import envelope as obspy_envelope
from scipy import signal as scipy_signal
from scipy.stats import kurtosis, skew
import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt
# ---------------------------------------------------------------------------
# Özellik çıkarımı — train_rf2.py ile BİREBİR AYNI
# ---------------------------------------------------------------------------
BANDPASS = (0.5, 15.0)
STA_S = 2.0
LTA_S = 20.0
MIN_NPTS = 200
MODEL_PATH = "rf_model.pkl"
def extract_features(tr) -> dict:
data = tr.data.astype(np.float64)
n = len(data)
df = tr.stats.sampling_rate
if n < MIN_NPTS:
return None
data -= np.mean(data)
sos = scipy_signal.butter(4, BANDPASS, btype="bandpass", fs=df, output="sos")
fdata = scipy_signal.sosfiltfilt(sos, data)
rms = np.sqrt(np.mean(fdata**2))
if rms < 1e-10:
return None
max_amp = np.max(np.abs(fdata))
peak_rms = max_amp / rms
zcr = ((fdata[:-1] * fdata[1:]) < 0).sum() / n
env = obspy_envelope(fdata)
env_smooth = scipy_signal.savgol_filter(env, min(51, n // 4 * 2 + 1), 3)
env_smoothness = np.std(np.diff(env_smooth)) / (np.mean(env_smooth) + 1e-10)
cum_e = np.cumsum(env ** 2)
idx90 = np.searchsorted(cum_e, 0.9 * cum_e[-1])
dur90 = idx90 / df
kurt_val = kurtosis(fdata)
skew_val = skew(fdata)
try:
cft = classic_sta_lta(fdata, int(STA_S * df), int(LTA_S * df))
sta_lta_max = np.max(cft)
sta_lta_mean = np.mean(cft)
triggers = trigger_onset(cft, 4.0, 1.5)
num_triggers = len(triggers)
except Exception:
cft = np.zeros_like(fdata)
sta_lta_max, sta_lta_mean, num_triggers = 0.0, 0.0, 0
nperseg = min(256, n // 2)
freqs, psd = scipy_signal.welch(fdata, df, nperseg=nperseg)
total_power = np.sum(psd) + 1e-30
def band_ratio(f_low, f_high):
mask = (freqs >= f_low) & (freqs < f_high)
return np.sum(psd[mask]) / total_power
low_ratio = band_ratio(0.5, 2.0)
mid1_ratio = band_ratio(2.0, 5.0)
mid2_ratio = band_ratio(5.0, 10.0)
high_ratio = band_ratio(10.0, 25.0)
dom_freq_idx = np.argmax(psd[freqs <= 25])
dom_freq = freqs[dom_freq_idx]
psd_norm = psd / total_power
spec_entropy = -np.sum(psd_norm * np.log(psd_norm + 1e-30))
spec_mean_freq = np.sum(freqs * psd_norm)
spec_bandwidth = np.sqrt(np.sum(((freqs - spec_mean_freq) ** 2) * psd_norm))
is_rf9f7 = 1 if "RF9F7" in (tr.stats.station or "").upper() else 0
feats = {
"peak_rms": peak_rms,
"zcr": zcr,
"kurt": kurt_val,
"skewness": skew_val,
"env_smoothness": env_smoothness,
"dur90_s": dur90,
"sta_lta_max": sta_lta_max,
"sta_lta_mean": sta_lta_mean,
"num_triggers": num_triggers,
"dom_freq": dom_freq,
"low_ratio": low_ratio,
"mid1_ratio": mid1_ratio,
"mid2_ratio": mid2_ratio,
"high_ratio": high_ratio,
"spec_entropy": spec_entropy,
"spec_bandwidth": spec_bandwidth,
"is_rf9f7": is_rf9f7,
}
# Grafik için ek veriler (feature dict'ine karışmaması için ayrı döndürülür)
return feats, fdata, cft, df
# ---------------------------------------------------------------------------
# Model yükleme (bir kere, global)
# ---------------------------------------------------------------------------
_bundle = None
def get_model():
global _bundle
if _bundle is None:
_bundle = joblib.load(MODEL_PATH)
return _bundle
def make_plot(fdata, cft, df, label, deprem_proba):
t = np.arange(len(fdata)) / df
fig, axes = plt.subplots(2, 1, figsize=(9, 5), sharex=True)
color = "#d64545" if label == "DEPREM" else "#3b6fd6"
axes[0].plot(t, fdata, color=color, linewidth=0.6)
axes[0].set_ylabel("Genlik (filtrelenmiş)")
axes[0].set_title(f"Tahmin: {label} (DEPREM olasılığı: %{deprem_proba*100:.1f})")
axes[1].plot(t, cft, color="purple", linewidth=0.8)
axes[1].axhline(4.0, color="black", linestyle=":", linewidth=1, label="Tetik eşiği")
axes[1].set_ylabel("STA/LTA")
axes[1].set_xlabel("Zaman (s)")
axes[1].legend(loc="upper right", fontsize=8)
plt.tight_layout()
return fig
def predict_mseed(file_obj, threshold):
if file_obj is None:
return "Lütfen bir MSEED dosyası yükleyin.", None, None
# Gradio 5.x'te gr.File genellikle doğrudan dosya yolu (str) döndürür,
# eski sürümlerde .name attribute'lu bir obje döndürüyordu — ikisini de destekle
filepath = file_obj if isinstance(file_obj, str) else getattr(file_obj, "name", None)
if not filepath:
return "Dosya yolu okunamadı.", None, None
try:
st = read(filepath)
except Exception as e:
return f"Dosya okunamadı: {e}", None, None
tr = None
for cha in ["EHZ", "HHZ", "BHZ", "EHN"]:
sel = st.select(channel=cha)
if len(sel) > 0:
tr = sel[0]
break
if tr is None:
tr = st[0]
result = extract_features(tr)
if result is None:
return f"Özellik çıkarılamadı (dosya çok kısa, min {MIN_NPTS} örnek gerekli).", None, None
feats, fdata, cft, df = result
bundle = get_model()
model, le, feature_cols = bundle["model"], bundle["le"], bundle["meta"]["feature_cols"]
X = np.array([[feats.get(c, 0.0) for c in feature_cols]])
X = np.where(~np.isfinite(X), 0.0, X)
proba = model.predict_proba(X)[0]
deprem_idx = list(le.classes_).index("DEPREM")
deprem_p = float(proba[deprem_idx])
label = "DEPREM" if deprem_p >= threshold else "NOISE"
emoji = "🔴 DEPREM" if label == "DEPREM" else "🔵 NOISE"
summary = (
f"## {emoji}\n\n"
f"**DEPREM olasılığı:** %{deprem_p*100:.1f}\n\n"
f"**NOISE olasılığı:** %{(1-deprem_p)*100:.1f}\n\n"
f"**Kanal:** {tr.stats.network}.{tr.stats.station}.{tr.stats.location}.{tr.stats.channel}\n\n"
f"**Başlangıç:** {tr.stats.starttime}\n\n"
f"**Süre:** {tr.stats.npts/tr.stats.sampling_rate:.1f} sn"
)
feat_table = "| Özellik | Değer |\n|---|---|\n"
for k, v in feats.items():
feat_table += f"| {k} | {v:.4f} |\n"
fig = make_plot(fdata, cft, df, label, deprem_p)
return summary, feat_table, fig
# ---------------------------------------------------------------------------
# Gradio arayüzü
# ---------------------------------------------------------------------------
with gr.Blocks(title="Sismik Sınıflandırma") as demo:
gr.Markdown(
"# 🌍 Sismik Sinyal Sınıflandırma (Marmara / RaspberryShake)\n"
"MSEED dosyası yükleyin, RandomForest modeli DEPREM veya NOISE olarak sınıflandırsın.\n\n"
"*Not: Model şu an sınırlı (Marmara bölgesi, RaspberryShake istasyonları) veriyle eğitildi, "
"gelişim aşamasındadır.*"
)
with gr.Row():
with gr.Column(scale=1):
file_input = gr.File(label="MSEED Dosyası (.mseed)", file_types=[".mseed", ".miniseed", ".seed"])
threshold_slider = gr.Slider(
minimum=0.1, maximum=0.9, value=0.5, step=0.05,
label="DEPREM karar eşiği",
info="Düşürürsen daha hassas (recall↑), yükseltirsen daha seçici (precision↑)"
)
submit_btn = gr.Button("Analiz Et", variant="primary")
with gr.Column(scale=1):
result_md = gr.Markdown(label="Sonuç")
with gr.Row():
plot_output = gr.Plot(label="Dalga Formu + STA/LTA")
with gr.Accordion("Detaylı özellik değerleri", open=False):
feat_output = gr.Markdown()
submit_btn.click(
fn=predict_mseed,
inputs=[file_input, threshold_slider],
outputs=[result_md, feat_output, plot_output],
api_name="predict",
)
gr.Markdown(
"---\n"
"**API kullanımı:** Bu Space'e programatik erişim için `gradio_client` kütüphanesini kullanabilirsiniz:\n"
"```python\n"
"from gradio_client import Client, file\n"
"client = Client(\"KULLANICI_ADI/SPACE_ADI\")\n"
"result = client.predict(file(\"ornek.mseed\"), 0.5, api_name=\"/predict\")\n"
"```"
)
if __name__ == "__main__":
demo.queue()
demo.launch(show_api=False)
|